Frequentist and Bayesian linear regression for large data sets. Useful
when the data does not fit into memory (for both frequentist and Bayesian regression),
to make running time manageable (mainly for Bayesian regression), and to reduce
the total running time because of reduced or less severe memory-spillover into
the virtual memory. This is an implementation of Merge & Reduce for linear regression
as described in Geppert, L.N., Ickstadt, K., Munteanu, A., & Sohler, C. (2020).
'Streaming statistical models via Merge & Reduce'. International Journal of
Data Science and Analytics, 1-17,